A method, storage medium and device for identifying earthquake faults

Abstract

The present invention discloses an earthquake fault identification method, storage medium, and device, which relate to the field of image processing. The method comprises: dividing a three-dimensional earthquake synthetic data set into a training set and a test set; constructing an earthquake fault identification model, including a Patch Embedding layer, an encoder, a multi-layer feature fusion module, a decoder, and a linear projection layer; the encoder and the decoder both include axial, spatial, and channel attention mechanisms. After passing through the Patch Embedding layer, the three-dimensional seismic data enters the encoder; the multi-layer feature fusion module fuses the feature map output by each encoding unit, and the output feature map of the multi-layer feature fusion module is input to the decoder and then passes through the linear projection layer to obtain the final fault identification result. The real earthquake data to be identified is input into the final earthquake fault identification model after training and testing to obtain the fault identification result. The present invention effectively improves the model's ability to identify small faults.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to an earthquake fault identification method, storage medium and equipment. Background Art

[0002] Earthquake fault identification is a key task in geological exploration and resource development, which directly affects the accuracy of mineral resource positioning, earthquake disaster prevention and geological structure research. Fault structure usually causes abnormal changes in seismic reflection signals, which appear as obvious lateral discontinuities on seismic images.

[0003] Traditionally, fault identification mainly relies on the visual interpretation of geological experts, which is inefficient and its accuracy is limited by subjective judgment, and it fails to fully utilize the multi-dimensional attributes of seismic data. Researchers have tried to use some attributes of earthquakes such as azimuth, coherence, curvature, etc. to assist in identification, but these methods are cumbersome and difficult to apply on a large scale. In order to improve processing efficiency and reduce reliance on manual interpretation, researchers have developed automatic and semi-automatic techniques such as phase unwrapping and ant tracking. However, due to the complexity of seismic data and the diversity of fault characteristics, these traditional methods still have great limitations in identification accuracy and efficiency.

[0004] In recent years, the rapid development of deep learning technology represented by convolutional neural networks has provided new solutions for earthquake fault identification. However, the traditional convolutional neural network structure lacks the ability to learn global context and long-distance spatial dependencies, which limits its performance in complex seismic data. At the same time, noise has a negative impact on its accuracy, making it difficult for the model to accurately identify faults in complex backgrounds. In order to solve these problems, many studies have tried to introduce attention mechanisms in recent years. Although the introduction of the attention mechanism has effectively improved the accuracy of fault prediction, there are still some key issues that need to be solved in small target fault identification:

[0005] (1) In earthquake fault identification, fault information often only occupies a small part of the image, while background information dominates. This imbalance causes the model to tend to favor background classification, thereby increasing the difficulty of identifying small faults.

[0006] (2) The identification of small faults often relies on the direction and structure of large faults as a guide. Since small faults are usually not prominent enough in seismic images, the model must have the ability to recognize multi-scale features in order to effectively utilize the information clues of large faults and accurately infer the existence and location of small faults.

[0007] (3) Although some models perform well in capturing long-range dependencies between patches, this often ignores the spatial local information within the earthquake patch. This practice may lead to a lack of attention to local details, thus affecting the overall recognition effect. Summary of the invention

[0008] The purpose of the present invention is to solve the problem of difficulty in identifying small faults and to propose a method for identifying earthquake faults, comprising the following steps:

[0009] S1, obtaining a 3D seismic synthetic data set and performing data preprocessing, and dividing the preprocessed data into a training set and a test set;

[0010] S2. construct an earthquake fault recognition model, which includes a Patch Embedding layer, an encoder, a multi-layer feature fusion module, a decoder, and a linear projection layer;

[0011] After being processed by the Patch Embedding layer, the 3D seismic data enters the encoder;

[0012] The encoder consists of multiple SCA blocks and convolutional downsampling layers to form multiple cascaded coding units;

[0013] The SCA block consists of a LayerNorm layer, an SCA module, and two convolutions. The input of the SCA block first passes through the LayerNorm layer. The output of the LayerNorm layer is input to the SCA module. The output of the SCA module is added to the input of the SCA block and then input to two cascaded convolutions. The input and output of the two cascaded convolutions are added to obtain the output of the SCA block.

[0014] The SCA module includes axial attention mechanism, spatial attention mechanism and channel attention mechanism;

[0015] The multi-layer feature fusion module fuses the feature maps output by each encoding unit;

[0016] The decoder consists of multiple SCA blocks and convolutional upsampling layers;

[0017] The output feature map of the multi-layer feature fusion module is input to the decoder, and the feature map output by the decoder passes through the linear projection layer to obtain the final fault recognition result.

[0018] S3. Training and testing the model to obtain a final earthquake fault identification model, inputting the real earthquake data to be identified into the final earthquake fault identification model to obtain a fault identification result.

[0019] Furthermore, S1 is specifically:

[0020] S11, constructing a 3D seismic synthetic data set by integrating multiple geological structures and randomized parameters, the data set including 3D seismic data and its corresponding labels;

[0021] S12, standardizing the data set, and applying random flipping, rotation and Gaussian filtering to perform data enhancement to obtain a preprocessed data set;

[0022] S13, divide the preprocessed data set into a training set and a test set in a ratio of 1:10.

[0023] Furthermore, the implementation method of the axial attention mechanism is as follows: project the 3D image X onto three orthogonal 2D planes to obtain three axial two-dimensional feature maps, perform linear transformation on the three axial two-dimensional feature maps to calculate the query and key vectors, while keeping the value V as the 3D image X, and calculate the attention map of each axis:

[0024]

[0025] in, represents the attention map of a certain axis, represents a query vector along a certain axis, represents the key vector of a certain axis, V is the value vector of the axis attention, and d is the size of each vector;

[0026] The attention matrices of each axis are weighted summed to generate the final axial attention map:

[0027] ,

[0028] in, Represents the final axial attention map, inline, xline, and time represent the three axes respectively. Represents the weight of the attention map of a certain axis;

[0029] The spatial attention mechanism uses three linear layers to compute the projection of the shared query, shared key, and spatial value, resulting in: , and ,in, , and Respectively represent their respective projection weights; and Projected to shape p The low-dimensional matrix of the spatial attention map is expressed as:

[0030] ,

[0031] in, represents the spatial attention map, represents the shared query vector, represents the shared key vector, represents a space-valued vector, Represents the projected , Represents the projected ;

[0032] The channel attention part uses the same and , and calculate the projection of the channel value ,in is the weight matrix of the channel value projection, and the channel attention map is defined as follows:

[0033] ,

[0034] in, represents the channel attention map, represents a channel value vector;

[0035] The output of the SCA module is represented as:

[0036]

[0037] in, represents the output of the SCA module, α and β represent weight coefficients, Express and Perform fusion operation.

[0038] Furthermore, the encoder consists of four cascaded coding units, the first three coding units are in turn cascaded by three cascaded SCA blocks and a convolutional downsampling layer, and the last coding unit consists of three cascaded SCA blocks;

[0039] The output of the Patch Embedding layer is input to the first encoding unit of the encoder, and the convolutional downsampling layer of the first encoding unit outputs the first feature map; the first feature map is input to the second encoding unit, and the convolutional downsampling layer of the second encoding unit outputs the second feature map; the second feature map is input to the third encoding unit, and the convolutional downsampling layer of the third encoding unit outputs the third feature map; the third feature map is input to the fourth encoding unit, and the fourth encoding unit outputs the fourth feature map.

[0040] Furthermore, the fusion method of the multi-layer feature fusion module is expressed as:

[0041] ,

[0042] in, represents the fusion features of the mth stage, , represents the i-th feature map, Represents the feature transformation function of the i-th feature map.

[0043] Furthermore, the decoder consists of four decoding units consisting of multiple SCA blocks and convolutional upsampling layers; the first three decoding units are sequentially composed of a convolutional upsampling layer and three cascaded SCA blocks, and the last decoding unit consists of a convolutional upsampling layer;

[0044] The fourth feature map is input into the convolution upsampling layer of the first decoding unit, and the output of the first decoding unit and the fusion feature After splicing, the convolutional upsampling layer of the second decoding unit is input, and the output of the second decoding unit and the fusion feature After splicing, the convolution upsampling layer of the third decoding unit is input, and the output of the third decoding unit and the fusion feature After concatenation, the convolutional upsampling layer is input into the fourth decoding unit to obtain the output of the decoder.

[0045] Furthermore, the loss function of the model training process is:

[0046]

[0047] Among them, L represents the loss function, N represents the total number of pixels, represents the true label of the i-th pixel, represents the predicted probability of the i-th pixel.

[0048] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned earthquake fault identification method is implemented.

[0049] The present invention also proposes an electronic device, comprising a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the above-mentioned earthquake fault identification method.

[0050] The beneficial effects brought by the technical solution provided by the present invention are:

[0051] The encoder and decoder of the model constructed based on the SCA module in this paper fully consider the multi-dimensional information and global context of seismic data by integrating axial, spatial and channel attention mechanisms (SCA modules). This design effectively overcomes the shortcomings of traditional convolutional neural networks in long-distance dependencies and complex backgrounds, and enhances the overall understanding of fault structures. A multi-level feature fusion module is used to perform multi-level fusion of the feature maps output by the encoder, and by combining shallow and deep features, the network is guided to focus on features of different scales. This design effectively improves the model's ability to recognize small faults and reduces the impact of noise. The model training adopts a mixed loss strategy, which can effectively solve the challenges of data imbalance and accurate fault identification in seismic data. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flow chart of an earthquake fault identification method according to an embodiment of the present invention;

[0053] Figure 2 is a structural diagram of an earthquake fault identification model according to an embodiment of the present invention;

[0054] Figure 3 This is a structural diagram of an SCA module according to an embodiment of the present invention;

[0055] Figure 4 is a block diagram of an electronic device in an exemplary embodiment of the present invention;

[0056] Figure 5 is the fault recognition result of the method of the embodiment of the present invention on the Kerry data set, wherein, Figure 5 (a) is a seismic image. Figure 5 (b) is the network recognition result of the model of the present invention. DETAILED DESCRIPTION

[0057] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0058] The flowchart of the earthquake fault identification method according to the embodiment of the present invention is as follows: Figure 1 , specifically including the following steps:

[0059] S1. Obtain a 3D seismic synthetic data set and perform data preprocessing, and divide the preprocessed data into a training set and a test set.

[0060] S11. A 3D seismic synthetic dataset is constructed by integrating multiple geological structures and randomized parameters. The dataset includes 3D seismic data blocks and their corresponding labels, ensuring a high diversity of geological features and simulation accuracy.

[0061] S12. Standardize the data set, and apply random flipping, rotation and Gaussian filtering to perform data enhancement to obtain a preprocessed data set.

[0062] S13, the preprocessed data set is divided into a training set and a test set in a ratio of 1:10. In the embodiment of the present invention, the data set includes 220 three-dimensional seismic data blocks and their corresponding labels, and the dimension of each data block is 128×128×128. The data set is divided into 200 training sets and 20 validation sets to ensure that the validation set contains enough representative samples to facilitate the performance evaluation of the subsequent model.

[0063] S2. Build an earthquake fault recognition model, which includes a Patch Embedding layer, an encoder, a multi-layer feature fusion module (MLFF), a decoder, and a linear projection layer.

[0064] like Figure 2 As shown, Figure 2 It is a structural diagram of the earthquake fault identification model of the embodiment of the present invention. The three-dimensional seismic data is processed by the Patch Embedding layer, divided into 4×4×4 blocks, and these small blocks are projected into a high-dimensional space of 16 channels. Next, the encoder part is entered. The encoder is composed of multiple SCA blocks and convolutional downsampling layers to form multiple cascaded encoding units. Each feature map output by the encoding unit integrates features of different scales through a multi-layer feature fusion module (MLFF) to enhance the recognition ability of small faults. The decoder is composed of multiple SCA blocks and convolutional upsampling layers, and the decoder also has a corresponding number of decoding units as the encoder. The multi-layer feature fusion module fuses the feature maps of each stage of the encoder to obtain fused feature maps of different scales. The output feature map of the multi-layer feature fusion module is input to the decoder, and the feature map output by the decoder passes through the linear projection layer to obtain the final fault identification result.

[0065] In the embodiment of the present invention, the encoder is composed of four cascaded coding units, the first three coding units are sequentially composed of three cascaded SCA blocks and a convolutional downsampling layer, and the last coding unit is composed of three cascaded SCA blocks.

[0066] The output of the Patch Embedding layer is input to the first encoding unit of the encoder, and the convolutional downsampling layer of the first encoding unit outputs the first feature map; the first feature map is input to the second encoding unit, and the convolutional downsampling layer of the second encoding unit outputs the second feature map; the second feature map is input to the third encoding unit, and the convolutional downsampling layer of the third encoding unit outputs the third feature map; the third feature map is input to the fourth encoding unit, and the fourth encoding unit outputs the fourth feature map.

[0067] The fusion method of the multi-layer feature fusion module is expressed as:

[0068] ,

[0069] in, represents the fusion feature of the mth stage, , represents the i-th feature map, represents the feature transformation function of the i-th feature map, which reduces the number of channels through 1×1 convolution and restores the resolution through upsampling. It is the feature map after the fusion of the first feature map, the second feature map, the third feature map, and the fourth feature map. It is the feature map after the fusion of the second feature map, the third feature map, and the fourth feature map. It is the feature map after the fusion of the 3rd feature map and the 4th feature map. It is the fourth feature map.

[0070] The decoder consists of four decoding units consisting of multiple SCA blocks and convolutional upsampling layers; the first three decoding units are sequentially composed of a convolutional upsampling layer and three cascaded SCA blocks, and the last decoding unit consists of a convolutional upsampling layer.

[0071] The fourth feature map is input into the convolution upsampling layer of the first decoding unit, and the output of the first decoding unit and the fusion feature After splicing, the convolutional upsampling layer of the second decoding unit is input, and the output of the second decoding unit and the fusion feature After splicing, the convolution upsampling layer of the third decoding unit is input, and the output of the third decoding unit and the fusion feature After concatenation, the convolutional upsampling layer is input into the fourth decoding unit to obtain the output of the decoder.

[0072] In the embodiment of the present invention, the SCA block is composed of a LayerNorm layer, an SCA module (spatial-channel-axial integrated attention module), and two convolutions. The input of the SCA block first passes through the LayerNorm layer, and the output of the LayerNorm layer is input to the SCA module. The output of the SCA module is added to the input of the SCA block and then input to two cascaded convolutions. The input and output of the two cascaded convolutions are added to obtain the output of the SCA block.

[0073] The SCA module includes axial attention mechanism, spatial attention mechanism and channel attention mechanism, refer to Figure 3 , Figure 3 4 is a structural diagram of the SCA module according to an embodiment of the present invention.

[0074] The implementation method of the axial attention mechanism is: the 3D image X ( ) is projected onto three orthogonal 2D planes to obtain three axial two-dimensional feature maps , , , linearly transform the two-dimensional feature map of the three axes to calculate the query and key vectors, while keeping the value V as the 3D image X, calculate the attention weight of each axis, and use broadcast multiplication with Calculate the attention map for each axis:

[0075] ,

[0076] in, represents the attention map of a certain axis, represents a query vector along a certain axis, represents the key vector of a certain axis, V is the value vector of the axis attention, and d is the size of each vector.

[0077] The attention matrices of each axis are weighted summed to generate the final axial attention map:

[0078] ,

[0079] in, Represents the final axial attention map, inline, xline, and time represent the three axes respectively. is a learnable parameter that represents the weight of the attention graph of a certain axis and satisfies .

[0080] The spatial attention mechanism uses three linear layers to compute the projection of the shared query, shared key, and spatial value, resulting in: , and ,in, , and Respectively represent their respective projection weights; and Projected to shape p A low-dimensional matrix, by After projection The transpose of is multiplied and the Softmax operation is performed to calculate the similarity between each feature and other spatial features; these similarities are Multiply them together to generate a spatial attention map, which is represented as:

[0081]

[0082] in, represents the spatial attention map, Represents the projected , Represents the projected .

[0083] The channel attention part uses the same and , captures the interdependence between feature channels and computes the projection of channel values ,in is the weight matrix of the channel value projection, and the channel attention map is defined as follows:

[0084] ,

[0085] in, represents the channel attention map.

[0086] The output of the SCA module is represented as:

[0087]

[0088] in, represents the output of the SCA module, α and β represent weight coefficients, Express and Perform fusion operation.

[0089] S3. Train and test the model to obtain a final earthquake fault identification model, input the real earthquake data to be identified into the final earthquake fault identification model to obtain a fault identification result.

[0090] The embodiment of the present invention selects the Adam optimizer, continuously adjusts the network parameters through the back propagation algorithm to reduce the training error, evaluates the model performance through cross-validation during the training process, and dynamically adjusts the hyperparameters to obtain the best training effect. At the same time, the learning rate warm-up and cosine annealing strategy are combined to deal with the oscillation phenomenon that may occur in the early stage of training.

[0091] The loss function of the model training process is:

[0092]

[0093] Among them, L represents the loss function, N represents the total number of pixels, represents the true label of the i-th pixel, represents the predicted probability of the i-th pixel.

[0094] In an exemplary embodiment, a computer-readable storage medium is included, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned earthquake fault identification method is implemented.

[0095] See also Figure 4In an exemplary embodiment, an electronic device is also included, including at least one processor, at least one memory, and at least one communication bus.

[0096] Wherein, a computer program is stored in the memory, and the computer program includes computer-readable instructions. The processor calls the computer-readable instructions stored in the memory through the communication bus to execute the above-mentioned earthquake fault identification method.

[0097] Figure 5 The fault recognition results of the method of the embodiment of the present invention on the Kerry dataset are shown, wherein: Figure 5 (a) is a seismic image. Figure 5 (b) is the network recognition result of the model of the present invention. Figure 5 (b) More small faults were successfully detected, demonstrating higher accuracy and comprehensiveness in complex geological environments.

[0098] Therefore, the present invention adopts the above-mentioned recognition method to achieve efficient and accurate fault recognition. This method reduces the sensitivity of the model to complex backgrounds and enhances the recognition ability of small target faults by integrating axial, spatial and channel attention mechanisms. The use of a multi-layer feature fusion module to combine information of different scales enables the model to make full use of multidimensional features and improve the overall recognition accuracy. In addition, by dynamically adjusting the learning rate and applying the early stopping strategy, the risk of overfitting is effectively reduced, making the application of the model on real data more reliable. The innovation of this method is that by optimizing the feature extraction and fusion process, the accuracy and robustness of earthquake fault recognition are significantly improved.

[0099] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying earthquake faults, characterized in that: The following steps are involved: S1, obtaining a 3D seismic synthetic data set and performing data preprocessing, and dividing the preprocessed data into a training set and a test set; S2. Constructing an earthquake fault recognition model, the model includes a Patch Embedding layer, an encoder, a multi-layer feature fusion module, a decoder, and a linear projection layer; After being processed by the Patch Embedding layer, the 3D seismic data enters the encoder; The encoder consists of multiple SCA blocks and convolutional downsampling layers to form multiple cascaded coding units; The SCA block consists of a LayerNorm layer, an SCA module, and two convolutions. The input of the SCA block first passes through the LayerNorm layer. The output of the LayerNorm layer is input to the SCA module. The output of the SCA module is added to the input of the SCA block and then input to two cascaded convolutions. The input and output of the two cascaded convolutions are added to obtain the output of the SCA block. The SCA module includes axial attention mechanism, spatial attention mechanism and channel attention mechanism; The multi-layer feature fusion module fuses the feature maps output by each encoding unit; The decoder consists of multiple SCA blocks and convolutional upsampling layers; The output feature map of the multi-layer feature fusion module is input to the decoder. The feature map output by the decoder passes through the linear projection layer to obtain the final fault recognition result; S3. Training and testing the model to obtain a final earthquake fault identification model, inputting the real earthquake data to be identified into the final earthquake fault identification model to obtain a fault identification result.

2. The earthquake fault identification method according to claim 1, characterized in that: S1 is specifically: S11, constructing a 3D seismic synthetic data set by integrating multiple geological structures and randomized parameters, the data set including 3D seismic data and its corresponding labels; S12, standardizing the data set, and applying random flipping, rotation and Gaussian filtering to perform data enhancement to obtain a preprocessed data set; S13, divide the preprocessed data set into a training set and a test set in a ratio of 1:

10.

3. The earthquake fault identification method according to claim 1, characterized in that: The implementation method of the axial attention mechanism is as follows: project the 3D image X onto three orthogonal 2D planes to obtain three axial two-dimensional feature maps, perform linear transformation on the three axial two-dimensional feature maps to calculate the query and key vectors, while keeping the value V as the 3D image X, and calculate the attention map of each axis: in, represents the attention map of a certain axis, represents a query vector along a certain axis, represents the key vector of a certain axis, V is the value vector of the axis attention, and d is the size of each vector; The attention matrices of each axis are weighted summed to generate the final axial attention map: ， in, Represents the final axial attention map, inline, xline, and time represent the three axes respectively. Represents the weight of the attention map of a certain axis; The spatial attention mechanism uses three linear layers to compute the projection of the shared query, shared key, and spatial value, resulting in: , and ,in, , and Respectively represent their respective projection weights; and Projected to shape p The low-dimensional matrix of the spatial attention map is expressed as: ， in, represents the spatial attention map, represents the shared query vector, represents the shared key vector, represents a space-valued vector, Represents the projected , Represents the projected ; The channel attention part uses the same and , and calculate the projection of the channel value ,in is the weight matrix of the channel value projection, and the channel attention map is defined as follows: ， in, represents the channel attention map, represents a channel value vector; The output of the SCA module is represented as: in, represents the output of the SCA module, α and β represent weight coefficients, Express and Perform fusion operation.

4. The earthquake fault identification method according to claim 1, characterized in that: The encoder consists of four cascaded coding units. The first three coding units are cascaded by three cascaded SCA blocks and a convolutional downsampling layer in turn, and the last coding unit is composed of three cascaded SCA blocks. The output of the Patch Embedding layer is input to the first encoding unit of the encoder, and the convolutional downsampling layer of the first encoding unit outputs the first feature map; the first feature map is input to the second encoding unit, and the convolutional downsampling layer of the second encoding unit outputs the second feature map; the second feature map is input to the third encoding unit, and the convolutional downsampling layer of the third encoding unit outputs the third feature map; the third feature map is input to the fourth encoding unit, and the fourth encoding unit outputs the fourth feature map.

5. A method for identifying earthquake faults according to claim 4, characterized in that: The fusion method of the multi-layer feature fusion module is expressed as: ， in, represents the fusion feature of the mth stage, , represents the i-th feature map, Represents the feature transformation function of the i-th feature map.

6. The earthquake fault identification method according to claim 5, characterized in that: The decoder consists of four decoding units consisting of multiple SCA blocks and convolutional upsampling layers; the first three decoding units are sequentially composed of a convolutional upsampling layer and three cascaded SCA blocks, and the last decoding unit consists of a convolutional upsampling layer; The fourth feature map is input into the convolution upsampling layer of the first decoding unit, and the output of the first decoding unit and the fusion feature After splicing, the convolutional upsampling layer of the second decoding unit is input, and the output of the second decoding unit and the fusion feature After splicing, the convolution upsampling layer of the third decoding unit is input, and the output of the third decoding unit and the fusion feature After concatenation, the convolutional upsampling layer is input into the fourth decoding unit to obtain the output of the decoder.

7. The earthquake fault identification method according to claim 1, characterized in that: The loss function of the model training process is: Among them, L represents the loss function, N represents the total number of pixels, represents the true label of the i-th pixel, represents the predicted probability of the i-th pixel.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the method according to any one of claims 1 to 7.

Images